Network 2025: Holiday Edition
What does networking look like over the next five years
What does networking look like over the next five years? Ok, any prognostication such as this would be lucky to be relevant after one year. However, people have to plan and strategize, so…
My first crack at this was Network 2025 and beyond. That remains the foundation of the thinking, but every six months is another six months of collecting data and testing thesis. So…
The Social Dilemma
The story starts at the top of the stack. For the cloud titans, the brand damage done during the 2016 election cycle, lingers, and 2020 was no exception with the release of “The Social Dilemma”. In 2019, Jeff Bezos shared a fair bit of his joint wealth, with his now ex-wife. By the end of 2020, he didn’t look much worse off for the experience. He still is doing, ok, financially.
Two things are true. The cloud titans provide enormous value to consumers, which is why they are so popular in the first place. Why so many people flock to them VOLUNTARILY. Secondly, we live in an era of activist/values-based buyers, and the wisdom of the crowd is fickle. Values-based capitalism is simple, in one sense. If you don’t like the way McDonald’s treats cows, don’t buy from McDonald’s. If you feel that one of the cloud titans has lost the plot, don’t use them. Voluntary transactions are the heart and soul of free markets. OTOH, sometimes we don’t always know how businesses are being run, and a little information is helpful.
Today there are community efforts, for example, DWEB, and then there are startups, for example, The@company These initiatives come either from a perspective that centralization is inherently an invitation to mischief, and therefore decentralization is the answer, and/or we need new approaches to information privacy and ownership, empowering consumers - these efforts go well beyond encryption, right to the heart of where information is stored, and who controls access to it.
It is hard to see this friction dissipating anytime soon and while many people who I know from my business life stand squarely behind Google’s handling of the Timnit Gebru exit, because their emotional reaction to put down the rebellion would have been the same, this is a year when Black Lives actually did matter for a short period of time, and people were rioting in the streets. Right and wrong, powerful and powerless, command and control, are not the issue here. The issue here is one more “black” eye for the cloud titans.
These black eyes are not terminal. The cloud titans provide enormous value which consumers and businesses voluntarily leverage, and some amount of bad press can be expected as a tax for having reached tall poppy status. None the less, this is friction at the top of the stack, and also entrepreneurial opportunity. The best strategies are ones that are a) compelling and b) unlikely to be followed by the incumbents. In this friction, I’m putting more money on entrepreneurs than I am on regulators.
Distributed Applications
Speaking of distributed applications, there is energy brewing all over the place here. Partly due to the cultural revolution against the social dilemma, where there are visions of applications, servers, and data being housed with the consumer. There is also big business jumping into 5G edge and IoT.
Edge cloud is a hot topic, but no one knows exactly where it is. Of course, it can be anywhere. It can be an NFL city availability zone, which Amazon recently announced there will be 20 more of next year. It can also be a telco central office, a hut, a strip mall, or a factory floor. All of these locations are candidates for distributed applications and investment/entrepreneurial energy.
Serverless/function as a service does not look like going mainstream anytime soon, partly because of the management complexity, but it does have some specific nice properties for stateless IoT transactions, though there are startups also working on stateful serverless. Serverless is a cool way to develop. It would be great to see it take off, but it does not appear to be ready for prime time just yet.
That leaves containers with all the current momentum. Virtual machines are still around and still have uses, but the momentum is with containers. The focus of distributed application solutions is likely to be on containers in the foreseeable future, but it will be interesting to see if IoT drives serverless into the conversation as well.
Network & Security Services
While Gartner penned the term “SASE” last year?, it seemed to really pick-up momentum in 2020 and is asking many challenging questions.
SD-WAN and SASE seem to be occupying different and overlapping spaces, and how this all sorts out is going to be an interesting thread to follow over the next few years.
For those occupying traditional network manager roles, SD-WAN is likely to be the solution leaned on to tackle the issues of expensive MPLS VPN services, routing across multiple telcos, leveraging multiple access methods (DSL, cable, Ethernet, TDM, WIFI, 4G/5G, and low-orbit satellite?), with application-aware intent. Need to get workers anywhere, with the best application performance, selectable from routes across multiple SP networks/peering locations, regardless of the application/service location? SD-WAN sounds like it might be a solution for that.
For those occupying traditional security manager roles, where a person - employee, partner, or guest, could be working on any kind of device, from anywhere in the world, and any time of the day and night, and where the security policy needs to be changed instantly, for example when you are pushing someone out the door, then SASE might be the right solution for that. Want a side of SD-WAN fries with that security solution? SASE is moving there too.
Putting aside the interesting and not inconsequential issues around SASE being an application-layer solution for any device, and SD-WAN being a network-layer solution, the big questions are:
Do networking people understand security?
Do security people understanding networking?
And do either understand distributed applications?
If the answer to all three of the above is “NO” then there might be quite a few global networks for managed experiences being built and operated. Will these all converge in the future? It seems probable. When? Probably not 2021, but who knows. It just takes someone with vision and a deep pocket to roll out of bed one day and decide it is a good idea.
While SASE and SD-WAN are what usually come to mind when people think about network as a service, the scenario that has grabbed my interest for a few years now is making network infrastructure better by leveraging cloud services. We are seeing some of this with cloud-managed solutions and SD-WAN, but I suspect there is much more to come, only limited by imagination.
The industry has seen Malware processing pop up in the cloud, and more recently, network analytics to realize improvements in the number of trouble tickets that have to be processed/resolved by humans. This is the dawning of the ML/AI era, with data coming from multiple networks, for the benefit of all networks. As we march towards autonomous networks, surely there is more to come.
Automation and Autonomy
Clearly, the DevOps era has given birth to the workflow taxonomy. There are so many repeatable things that have to move through a pipeline to get to some predicted/intended outcome. This is the heart of automation. There are many technology suppliers coming at this from different angles, in addition to DIYers.
Autonomy, on the other hand, can be characterized as automation in the sense that it is a non-human response, but it is differentiated by learning and response to events and conditions that are neither scheduled, expected, or initiated, in the same way as a workflow is.
While the prospect of completely autonomous networks is hard to imagine by 2025, the signs of progress are already emerging, and there will surely be more over the next five years. It is on this dimension, that technology leadership in networking may well be judged over the next 5-10 years.
Routing control planes already take on an ambitious goal, and complexities aside, execute that goal fairly well. However, they are limited by information - the industry is cautious about putting too much information into them in case that leads to failure; limited by CPU cycles/cores, and limited by memory.
The industry is at the early stages of extracting and leveraging information from a much broader set of sources. Real-time telemetry is expected to increase the amount of information used by ML, but there are other sources of operational information as well, laying around, waiting to be leveraged in value-added ways.
Automated anomaly identification and resolution are at the forefront of this trend. Automating the establishment of baselines, detecting when there are variants, and providing information about the variants that assist network managers in doing root cause analysis. And when there are not any variants, this automated monitoring/insight, provides faster time to “proof of innocence” - no, it is not the network’s fault, everything is as it normally is, the problem with the application is somewhere else.
Some systems are already demonstrating a level of automated troubleshooting and resolution, in problem-domain-constrained parts of the network, where deep knowledge can be learned and applied.
ML anomaly identification and action will likely evolve to proactive routing (speculation) - changing routes just because a problem looks like it is building, rather than waiting for the problem to actually disrupt routing. Proactive routing will likely be another example of augmented routing.
While it is unlikely that something as complex as a large-scale WAN network and the Internet as a whole can be fully autonomous by 2025, it is going to be of tremendous interest to see just how much progress the industry can make by 2025.
Overlays
Remember when there was just frame relay or routing. Then there was VPNs, where we discovered more than routing had to be virtualized. Now, the overlay is a full-blown thing.
BGP-based IPVPNs have led to BGP-based Ethernet VPNs. The benefits of using BGP for both seems clear:
In addition to BGP-based “overlays” in the VPN sense, BGP-LS is also being touted. Over the last decade, Contrail, Nicira/NSX, etc reminded us that a network does not end at an Ethernet switch port, but goes all the way to the application server. Contrail Networking/Tungsten fabric is also BGP-based.
At the same time, the somewhat complicated to understand CenturyLink outage reminded us that the problem of connecting networks may not yet be a solved problem / requires some more work.
Overlay considerations are now important design/architecture considerations. Overlays are a thing, and they are pushing buttons on overlay vs underlay technology decisions/choices.
The Overlay conversation over the next 5-10 years is going to be an interesting one, whether it is the network access to network access/application overlay or the server-to-server overlay; the latter particularly picking up steam as edge cloud/distribution applications/cloud-native gain momentum.
The Underlay
Ahh yes. The hard-working, unappreciated, workhorse of networking, moving packets around at the speed of light, without barely a complaint.
Ok, maybe that is a little exaggeration.
We are now entering a period of what seems like growing adoption of Segment Routing, for at least new builds, and where 5G may play prominently.
And yet, it seems like there is still much to learn about segment routing. Was state simply moved from the control plane to the data plane, or was there a net loss in information, leading to a loss in capability?
If capability was lost (putting aside policy-based recovery of that capability), is the loss terminal or inconsequential?
Going back a few decades, Frame Relay was this technology that everyone just kind of loved. A simple approach to a WAN data link layer, no PID-less underlay leaning on signaled tunnels, just boring old RFC1294 (that no doubt later inspired 1483), and then ATM came along. ATM had all these QoS capabilities, and PNNI-based routing capabilities, and yet, today, it barely rates a mention. It just had a bunch of stuff that was not commonly used, even though very useful in specific use cases.
RSVP-TE MPLS is beginning to feel like ATM. RSVP-TE is widely deployed, many network managers are familiar and comfortable with it, it creates tunnels in an effective way, it is still part of the conversation, and yet, it just feels like it has a bunch of capabilities that most network managers are not using, and that is perhaps its biggest weakness.
The classic Clay Christensen definition of disruptive innovation is a market with well-established incumbents, delivering way more functionality than can be consumed by customers, and new companies entering the market with less functionality and lower prices.
This is the industry dynamic currently playing out in the underlay, arguably energized by hyperscalers, but nonetheless, potentially inevitable. In the next 3-5 years, this will play out. We will learn things about Segment Routing that we do not currently know. We will learn or re-learn what capabilities different customer segments want in the underlay. This process is going to consume significant industry cycles.
Silicon Wars
Without question, one of the more interesting announcements this year was Cisco’s announcement to take on Broadcom in the “merchant” silicon space. Cisco knows a little bit about networking and has built its own silicon for many years.
Put a gun against my head and force me to make a prediction, and I would likely predict that Broadcom will prevail. However, that is not a comfortable prediction. Both Broadcom and Cisco have huge financial muscle and can take on big fights. Cisco has to learn/execute a new business model and balance the interests of a systems business and a chip business. Here Broadcom has clear advantages.
Of all the technical angles that could be taken on this battle of the titans, the one that interests me the most, today, is around chip APIs (which Cisco has to build a muscle around supporting). Cisco has thrown a thesis out that one API for a family of silicon is much better for Cisco’s direct customers (for example hyperscalers) and potential OEM customers/competitors (other network equipment suppliers).
If you have ever worked in a shop that had to suddenly change the roadmap because it became clearer what the implications of moving from one Broadcom chip to another can be, especially given Broadcom’s acquisitions, then this is not a lightweight issue; it is an issue that deserves a test in the marketplace, and Cisco is looking to provide that test. Let’s see what happens…
Optical Wars
Speaking of Cisco, Cisco’s acquisition of Acacia, if it ever closes, has many speculating about the consequences for industry change. While we all wait, like anxious children around a Christmas tree, Cisco is feeding us the occasional teaser to whet our appetites:
One part of the wait is knowing with more certainty whether post-acquisition, Cisco will supply optics to others, and who. Any intent they express prior to closing the Acacia acquisition is likely to meet with a little skepticism, whether warranted or not.
However, that’s not the issue for which the next 5-10 years of networking will be mostly concerned with. The issue the industry waits for an answer on is how deeply will pluggable optics cut into the optical systems business.
Packet and Optical suppliers have been teasing us with the specter of this showdown for 10+ years, and now, with 400ZR being rolled out, it looks like the showdown is finally coming. Yes, optical systems companies may have a play with their own optical pluggables, but the long-term issue is whether WANs start to look a little more like LANs, with the optical layer just a transmission substrate, and not a distinct network layer.
At this time, that reality seems unlikely, end to end, by 2025, as optical specialists start deploying 800 Gbps per wavelength capabilities, and spectral efficiency remains an important consideration in many people’s eyes, as does DWDM. In addition, pluggables are generally considered limited in reach, especially the standardized flavors. Proprietary may be a different issue. So there appears today, a higher probability of a shakeup in metro links than long haul routes.
While Router vendors have the challenge of putting forth an optical solution that is appealing, optical systems vendors have the challenge of making a multi-layer solution as easy to manage/orchestrate as a single layer solution.
Without a doubt, how this plays out by 2025 is going to be of great interest, and metro connectivity could be in for a real shakeup, especially if multiple router suppliers can execute on solutions that include passive optical components.
Other
Black Swans cannot be predicted, by definition, so calling this section “Black Swans” would have been intellectually convoluted. However, there are things floating below the surface that are just waiting for a wave of interest to pick them up and move the needle on. Probabilities are low, but wild speculation should always be encouraged :-)
Null Safe Internet
Quality is a constant in networking. It is important. It waxes and wanes. Quality definitely impacts technology supplier pricing, network customer experience, and whether there is time in the day to work on other stuff. It is tempting to assert that it is the most important thing. However, the Internet works fairly well, most of the time, despite mixed levels of quality, precisely because of its underlying principles, assumptions, and survivability.
Nonetheless, as the world of programming languages moves in the direction of null safety, it is not unreasonable to speculate on how and when the networking industry will pick up on this trend. Null errors are not the only issue with respect to quality, but code that is prone to falling over because of nulls is a pain in the neck and requires defensive coding which adds complexity and reduces readability. All this means more overhead/tax and less value-adding capabilities.
While it is hard to imagine this being the primary thrust in future market disruptions, it could be part of an overall disruption approach. It is not a high probability, but it is certainly laying around for someone to pick it up and run with it.
Null safe Internet is a proxy for all issues related to quality.
5G—>6G
At this time, it is hard to imagine 6G even being fully standardized by 2025, let alone being deployed. However, there are already vendors touting their position as leaders in 6G standardization and research, so…In addition, with 5G radios being deployed ahead of 5G packet cores, it does feel like 5G is going to be a slow roll and not a big bang event. Potentially leaving 6G as the generation where the most notable transformations are realized, potentially use cases that were originally promised for 5G.
Current predictions for 6G are pilots in 2026 (South Korea), deployments in 2028, and mass commercialization in 2030.
The situation with Open RAN is not clear today. Some news days have optimistic stories, others pessimistic. There is no real clarity about Open RAN in 5G, so it may be one of the major realizations of 6G.
More generally, mobility may be the last bastion of Telco-specific standards and technology. These technologies may suffer from being subscale, and so 6G may be an opportunity to dig into interesting questions on WIFI’s role in telco networks, what IoT really needs, and overall network design/architecture.
Another interesting question to resolve over the next five years is wireline/wireless infrastructure convergence. From the perspective of the Routing community, it makes all kinds of sense, especially if the mid/backhaul sections of the network are not mobility specific. Unfortunately, it does appear like front haul will continue to have mobility-specific complexities in the near future. With the ongoing desire to always support multiple generations of radio/mobility, that may not change any time soon.
Open CLI
Having an open, standardized CLI is unlikely to be at the top of anyone’s todo list. Especially in an era of increasing programmability and cooperative processing between network nodes and controllers/automation systems.
This is one of those things lying around for someone to pick up and run with. Perhaps SONiC will turn out to be the trojan horse for establishing this? Maybe one of the other independent routing stacks will. The bottom line though is that as Yang-models become more common / used, and python continues to grow as the go-to tech for how CLIs are built, in part or in full, this opportunity seems to be ripe.
Again, low probability, but just sitting around for someone/some force to sweep up if there is interest and energy.
Quantum Networking
Quantum Networking does not seem poised to be disruptive by 2025. However, it does feel like quantum innovation is accelerating, and there are some use cases that will emerge alongside, but not replacing classic networking (the same is true for quantum computing and classic computing).
As the pace of innovation is accelerating, this is something to check-in on occasionally, just to know where it is at. I know I have some more reading to do here, and I suspect many others do as well.
IGP Policy Engines
Typically, BGP policy engines have been an area of differentiation between the major router vendors. IF the information loss experienced by going to Segment Routing results in a lack of capability that is later desired, there is the potential for policy engines to become important as a way of injecting information back into the network. This is theoretical speculation only, with no practical basis in fact. We will just have to see how this plays out.
IPv6
Not surprisingly, APAC seems to be leading the charge towards SRv6/IPv6.
Will SRv6 and/or IPv6 dominate networking by 2025? That seems unlikely at this time, but it cannot be completely dismissed. IPv6 usage has risen to an interesting level, and with what could be speculated as a declining interest in complex per-hop QoS, there is the possibility of a few silicon gates laying around looking for something to do. The silicon dividend could equally result in AI inference engines. What a silicon dividend may result in is perhaps not well known at this time, but SRv6 / IPv6 is on the list of potential benefactors.
That technology perspective aside, it is ultimately up to operators to assess whether or not they need to move to IPv6. Some already have. Others will make decisions by 2025.
ARM & RISC-V
As attention remains fixed on Cisco vs Broadcom, ARM continues to make waves outside of networking, and who knows what impact it will have inside networking by 2025.
Today, there are a number of access routers using ARM-based processors for routing protocols. There are for sure complaints that these options seem underpowered and do on occasion limit capabilities. On the other hand, Apple’s pivot to ARM in its laptops has shone a spotlight on the potential for dramatically improved performance with lower power requirements. ARM may also be interesting in Edge Cloud compute scenarios.
At the same time, RISC-V is getting more press. RISC-V is an open-sourced instruction set that some see as the ascendent replacement for the current ARM architecture. ARM is of course licensed.
Static Routes
It seems like the exception these days, that I watch a network design discussion on YouTube, where the lowly static route is not put forth as the solution to some ill. My last prediction is that in 2025, there will still be static routes ;-) This is not in the black swan category, this is a given :-)
Happy Holidays!
